Pipeline Defects Risk Assessment Using Machine Learning and Analytical Hierarchy Process

Abstract

Pipelines are the most important way to transport large amounts of dangerous substances as oil and gas, through long distances, due to their advantages in terms of safety and low cost. However, failures and leaks in pipelines may happen and sometimes they generate catastrophic consequences. In this paper we propose an approach for the risk assessment of oil and gas pipeline defects leveraging machines learning algorithms and multi-criteria decision methods (MCDM), with the objective of accompanying decision-makers for prioritizing risk mitigation activities. The pipeline defects risk assessment approach proposed is based on some machines learning algorithms, which allows to cluster ILI (In Line Inspection) data performed by smart pigs in a group of clusters by using K-means method, then, two classifications methods (decision trees and neural network) are applied on clusters in order to construct a classification model of defects risk on pipe in three level (High, Medium and Low) according to theirs criticizes. The discovered models are assessed using cross validation, which allows choosing a model based on a decision tree as a pipeline defects risk classification and prediction model. For scheduling maintenance and reparation operations we apply the multi-criteria decision method AHP (Analytical Hierarchy Process) in order to rank-order defects which belong to the High class according to theirs criticizes degree.